U.S. patent application number 14/004002 was filed with the patent office on 2014-01-09 for sar data processing.
This patent application is currently assigned to ASTRIUM LIMITED. The applicant listed for this patent is Alexander Walker Wishart. Invention is credited to Alexander Walker Wishart.
Application Number | 20140009326 14/004002 |
Document ID | / |
Family ID | 44247017 |
Filed Date | 2014-01-09 |
United States Patent
Application |
20140009326 |
Kind Code |
A1 |
Wishart; Alexander Walker |
January 9, 2014 |
SAR DATA PROCESSING
Abstract
An apparatus is disclosed for a spaceborne or aerial platform
having a frequency demultiplexer for frequency demultiplexing a
signal corresponding to a range line or an azimuth line of SAR
data, and including information about a plurality of target points,
into a plurality of frequency channels, and a compression device
for performing compression on each frequency channel, each
frequency channel signal having information about the same target
points. The frequency demultiplexer and the compression device can
be implemented in hardware. The apparatus may be used for either or
both of the range compression and the azimuth compression of a SAR
arrangement on board a spaceborne or aerial platform and the SAR
arrangement may generate a plurality of sub-images corresponding to
the frequency channels from the SAR raw data. The sub-images may be
combined by averaging in order to reduce the volume of memory
required to store the SAR data.
Inventors: |
Wishart; Alexander Walker;
(Stevenage, GB) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Wishart; Alexander Walker |
Stevenage |
|
GB |
|
|
Assignee: |
ASTRIUM LIMITED
Stevenage, Hertfordshire
GB
|
Family ID: |
44247017 |
Appl. No.: |
14/004002 |
Filed: |
March 9, 2012 |
PCT Filed: |
March 9, 2012 |
PCT NO: |
PCT/EP2012/054164 |
371 Date: |
September 23, 2013 |
Current U.S.
Class: |
342/25D |
Current CPC
Class: |
G01S 13/90 20130101;
G01S 13/9011 20130101 |
Class at
Publication: |
342/25.D |
International
Class: |
G01S 13/90 20060101
G01S013/90 |
Foreign Application Data
Date |
Code |
Application Number |
Mar 10, 2011 |
EP |
11275041.9 |
Claims
1. Apparatus for a spaceborne or aerial platform, comprising: a
frequency demultiplexer for frequency demultiplexing a signal
corresponding to a range line or an azimuth line of Synthetic
Aperture Radar (SAR) data, and containing information about a
plurality of target points, into a plurality of frequency channels,
each frequency channel signal containing information about the
plurality of target points; and compression means for performing
compression on each frequency channel.
2. Apparatus according to claim 1, wherein the frequency
demultiplexer and the compression means are implemented in
hardware.
3. Apparatus according to claim 2, wherein the compression means
comprises: one or more compression filters.
4. Apparatus according to claim 3, wherein the one or more
compression filters comprise: a compression filter for providing an
output signal to the frequency demultiplexer, the compression
filter providing a filtering function with an imaged response for
each frequency channel.
5. Apparatus according to claim 4, wherein a frequency response of
the compression filter is configured for providing the imaged
response as H'(e.sup.i.omega.)=H(e.sup.i.omega.K), where
H(e.sup.i.omega.) is a frequency response of a prototype filter
h(n) and: h ' ( n ) = { h ( n / K ) , n = 0 , .+-. K , .+-. 2 K 0 ,
otherwise ##EQU00005## and where K is the number of frequency
channels.
6. Apparatus according to claim 5, wherein the frequency
demultiplexer comprises: block fine processing stages and block
coarse processing stages, the block fine processing stages being
configured to generate from the signal corresponding to a range
line or an azimuth line one or more first frequency division
multiplex FDM signals with a number of even index channels and one
or more second FDM signals with a number of odd index channels, a
region of a spectrum corresponding to the odd index channels having
been nulled out in the one or more first FDM signals and a region
of the spectrum corresponding to the even index channels having
been nulled out in the one or more second FDM signals, the one or
more first FDM signals and second FDM signals being provided to
separate coarse processing stages and each coarse processing stage
generating a plurality of narrow band channels from its respective
FDM signal, each block coarse processing stage being configured for
performing a filtering function with response transition bands
which lie in the nulled out regions of the spectrum of the
associated FDM signal.
7. An arrangement for creating a SAR image on a spaceborne or
aerial platform in combination with the apparatus of claim 1,
comprising: analogue to digital converter means for converting
received SAR echoes into digital samples; a processor having range
processing means for carrying out range compression of arrays of
samples corresponding to range lines and azimuth processing means
for carrying out azimuth compression on arrays of samples
corresponding to azimuth lines, wherein at least one output of the
range compression means and the azimuth compression means includes
the apparatus, the processor being configured for generating a
plurality of sub-images corresponding to the frequency channels
into which a signal corresponding to a range line or an azimuth
line is demultiplexed; and image generator means for combining the
sub-images.
8. An arrangement for creating SAR images on a spaceborne or aerial
platform in combination with the apparatus of claim 1, comprising
analogue to digital converter means for converting received SAR
echoes into digital samples; a processor having range processing
means for carrying out range compression of arrays of samples
corresponding to range lines and azimuth processing means for
carrying out azimuth compression on arrays of samples corresponding
to azimuth lines, wherein both the range processing means and the
azimuth processing means include the apparatus, wherein the
frequency demultiplexer in the range processing means is configured
to demultiplex a signal corresponding to a range line into
K.sub.range frequency channels and the frequency demultiplexer of
the azimuth processing means is configured to demultiplex a signal
corresponding to an azimuth line into K.sub.azimuth frequency
channels to produce K.sub.range K.sub.azimuth frequency channels,
the processor being configured to generate K.sub.range
K.sub.azimuth sub-images corresponding to the K.sub.range
K.sub.azimuth frequency channels; and image generator means for
combining the sub-images.
9. An arrangement according to claim 7, wherein the image generator
means is configured to average the sub-images to provide a SAR
image.
10. An arrangement according to claim 7, comprising; a receiver
configured to receive echoes from two or more SAR antennas, the
image generator means being configured to generate an interferogram
from the sub-images.
11. The arrangement of claim 7 in combination with a satellite.
12. A method of generating a SAR image on a spaceborne or aerial
platform, comprising receiving echoes of a chirp transmitted from a
SAR antenna; digitising the echoes into digital samples; carrying
out range compression on arrays of samples corresponding to range
lines; carrying our azimuth compression on arrays of samples
corresponding to azimuth lines, wherein at least one out of
carrying out range compression and carrying out azimuth compression
includes frequency demultiplexing signals corresponding to the
arrays into a plurality of channels, each signal corresponding to
an array containing information about a plurality of target points
and each frequency channel containing information about said
plurality of target points, and carrying out image formation on
each channel such that the range compression and the azimuth
compression generates a plurality of sub-images corresponding to a
plurality of frequency channels; and combining said sub-images.
13. A method according to claim 12, wherein the image formation of
each channel is carried out before the frequency demultiplexing of
the signals and the frequency demultiplexing comprises: frequency
demultiplexing the frequency channels in compressed form.
14. A method according to claim 13, wherein the range compression
comprises: frequency demultiplexing the range lines into a
plurality of channels and range compressing the channels
individually using a compression filter to form a plurality of
intermediate matrices, and wherein the azimuth compression includes
frequency demultiplexing each azimuth line of each intermediate
matrix into a plurality of frequency channels and azimuth
compressing each frequency channel individually using a compression
filter to form the plurality of sub-images.
15. A method according to claim 13, wherein the range compression
comprises: frequency demultiplexing the range lines into a
plurality of channels and range compressing the channels
individually using a compression filter to form a plurality of
intermediate matrices, and wherein the azimuth compression includes
averaging the samples of each azimuth line to form the plurality of
sub-images.
16. A method according claim 12, wherein combining said images
comprises: averaging the sub-images to produce a compressed image,
or wherein receiving echoes from an antenna comprises receiving
echoes from a first antenna and combining the sub-images includes
correlating corresponding sub-images from the first and a second
antenna to form a number of sub-interferograms and averaging the
sub-interferograms to generate a compressed interferogram.
17. Apparatus according to claim 1 wherein the compression means
comprises: one or more compression filters.
18. Apparatus according to claim 1, wherein the frequency
demultiplexer comprises: block fine processing stages and block
coarse processing stages, the block fine processing stages being
configured to generate from the signal corresponding to a range
line or an azimuth line one or more first frequency division
multiplex FDM signals with a number of even index channels and one
or more second FDM signals with a number of odd index channels, a
region of a spectrum corresponding to the odd index channels having
been nulled out in the one or more first FDM signals and a region
of the spectrum corresponding to the even index channels having
been nulled out in the one or more second FDM signals, the one or
more first FDM signals and second FDM signals being provided to
separate coarse processing stages and each coarse processing stage
generating a plurality of narrow band channels from its respective
FDM signal, each block coarse processing stage being configured for
performing a filtering function with response transition bands
which lie in the nulled out regions of the spectrum of the
associated FDM signal.
19. An arrangement according to claim 8, wherein the image
generator means is configured to average the sub-images to provide
a SAR image.
20. A method according to claim 12, wherein the range compression
comprises: frequency demultiplexing the range lines into a
plurality of channels and range compressing the channels
individually using a compression filter to form a plurality of
intermediate matrices, and wherein the azimuth compression includes
frequency demultiplexing each azimuth line of each intermediate
matrix into a plurality of frequency channels and azimuth
compressing each frequency channel individually using a compression
filter to form the plurality of sub-images.
Description
FIELD OF THE INVENTION
[0001] The invention relates to processing of synthetic aperture
radar (SAR) data.
BACKGROUND OF THE INVENTION
[0002] Synthetic aperture radar can be used to generate high
resolution images of a target. The target is illuminated by pulses
of radiation and the echoes of the pulses are detected and
processed to form image data. SAR can be used on a spacecraft, such
as a satellite. The raw echo data in digital form is typically
referred to as a Level 0 product. SAR images formed from the Level
0 products are typically referred to as a Level 1 product.
[0003] Typically, the Level 0 raw data obtained by SAR on
satellites is processed on ground into a Level 1 product and the
satellites therefore have to store the Level 0 data until it can be
transmitted to ground. This can be a problem when the satellite
does not frequently have the opportunity to communicate with a
ground station. For example, on planetary missions, the satellites
would have to store received data for an extended time. Moreover,
satellites such as low Earth orbit (LEO) satellites would have to
store data when they are not in communication with a ground
station. The number of images that can be obtained is then limited
by the available storage on the satellite.
[0004] It is sometimes desired to have more than one SAR antenna on
a satellite in order to, for example, generate interferometric
images or complex coherence functions. By increasing the number of
SAR antennas, the amount of Level 0 raw data generated is also
increased, exacerbating the problem.
[0005] It has been proposed that there may be advantages to
carrying out image processing on the satellite. For example, if
lower resolution images are acceptable, the level 0 data could be
processed into lower resolution images on the satellite and thereby
allow the satellites to store data for a larger number of images
than if the stored data was Level 0 data. Moreover, in some
applications it is desired to provide real-time or near real-time
monitoring of a target scene and it may take too long time to send
the image data to ground for processing.
[0006] State of the art SAR image processing is implemented in
software on ground. Currently available CPUs suitable for space
on-board processing would be too slow to run the computationally
intensive software used on ground effectively. This is because the
processing rate of fault tolerant general purpose CPUs suitable for
space on-board processing is orders of magnitude less than
terrestrial systems.
[0007] The invention aims to address the above and other
problems.
SUMMARY OF THE INVENTION
[0008] According to the invention, there is provided an apparatus
for a spaceborne or aerial platform, comprising a frequency
demultiplexer for frequency demultiplexing a signal corresponding
to a range line or an azimuth line of SAR data and comprising
information about a plurality of target points into a plurality of
frequency channels, each frequency channel signal comprising
information about the plurality of target points, and compression
means for performing image formation on each frequency channel.
[0009] Since the data is frequency demultiplexed into a number of
channels, the bandwidth of the channels for which compression is
carried out is narrower and the compression can be carried out in
hardware. Both the frequency demultiplexer and the compression
means can be implemented in hardware.
[0010] For this and other reasons, the invention makes it possible
for the range compression and azimuth compression to be carried out
on a spaceborne or aerial platform. For example, the invention
makes it possible for the range and azimuth compression to be
carried out on a satellite or other spacecraft. The invention also
makes it possible for the range or azimuth compression to be
carried out an aerial vehicle, for example, an unmanned aerial
vehicle (UAV).
[0011] The invention combines multi-rate digital signal processing
techniques in a novel way to minimise the implementation complexity
of the SAR image generation.
[0012] The compression means may comprise one or more compression
filters.
[0013] The output of the frequency demultiplexer may be provided to
the inputs of a plurality of compression filters. The frequency
demultiplexer isolates the frequency channels and each frequency
channel may be provided to a respective compression filter.
[0014] Alternatively, the one or more compression filters may
comprise a compression filter that provides an output signal to the
frequency demultiplexer, the compression filter h'(n) providing a
filtering function with an imaged response for each frequency
channel. The frequency response of the compression filter providing
the imaged response may be H'(e.sup.i.omega.)=H(e.sup.i.omega.K),
where H(e.sup.i.omega.) is the frequency response of a prototype
filter h(n) and
h ' ( n ) = { h ( n / K ) , n = 0 , .+-. K , .+-. 2 K 0 , otherwise
##EQU00001##
and where K is the number of frequency channels.
[0015] The frequency demultiplexer then isolates the frequency
channels in compressed form.
[0016] By rearranging the order of the frequency demultiplexing and
the compression and carrying out the compression function before
the demultiplexing function, the apparatus can be even more
computationally efficient.
[0017] The frequency demultiplexer may comprise a block
channeliser. The block channeliser may comprises block fine
processing stages and block coarse processing stage, the block fine
processing stages being configured to generate from the signal
corresponding to a range line or an azimuth line one or more first
frequency division multiplex FDM signals with a number of even
index channels and one or more second FDM signals with a number of
odd index channels, the region of the spectrum corresponding to the
odd index channels having been nulled out in the one or more first
FDM signals and the region of the spectrum corresponding to the
even index channels having been nulled out in the one or more
second FDM signals. The one or more first FDM signals and second
FDM signals may be provided to separate coarse processing stages
and each coarse processing stage may generate a plurality of narrow
band channels from its respective FDM signal. Each block coarse
processing stage may perform a filtering function with a response
the transition bands of which lie in the nulled out regions of the
spectrum of the associated FDM signal.
[0018] According to the invention, there is also provided an
arrangement for creating a SAR image on a spaceborne or aerial
platform, comprising: analogue to digital converter means for
converting received SAR echoes into digital samples; a processor
comprising range processing means for carrying out range
compression of arrays of samples corresponding to range lines and
azimuth processing means for carrying out azimuth compression on
arrays of samples corresponding to azimuth lines, wherein at least
one out of the range compression means and the azimuth compression
means comprises the apparatus defined above, the processor being
configured to generate a plurality of sub-images corresponding to
the frequency channels into which a signal corresponding to a range
line or an azimuth line is demultiplexed; and image generator means
for combining the sub-images.
[0019] Both the range processing means and the azimuth processing
means may comprise said apparatus having a frequency demultiplexer
and compression means, wherein the frequency demultiplexer in the
range processing means is configured to demultiplex a signal
corresponding to a range line into K.sub.range frequency channels
and the frequency demultiplexer of the azimuth processing means is
configured to demultiplex a signal corresponding to an azimuth line
into K.sub.azimuth frequency channels such that the processor
generates K.sub.range K.sub.azimuth sub-images.
[0020] The image generator means may be configured to average the
sub-images to provide a SAR image. The volume of memory required to
store a multi-looked, averaged image generated from the raw data
would typically be significantly smaller than the volume of memory
required to store the raw high resolution data. By averaging the
sub-images, significantly smaller memory is required to store the
data to be transmitted to ground and data for a larger number of
images can be stored in between transmissions to the ground.
Additionally, by carrying out image generation on the satellite,
real-time, or at least quicker, analysis of the imaged target can
be carried out. The target may be monitored and alerts may be
raised in response to changes to the images generated on the
satellite.
[0021] The receiver may be configured to receive echoes from two or
more SAR antennas and the image generator means may be configured
to generate an interferogram or a complex coherence map from the
sub-images.
[0022] According to the invention, there is also provided a
satellite comprising the above defined apparatus or arrangement.
The satellite may also comprise means for transmitting the
processed SAR data in a downlink.
[0023] According to the invention, there is also provided a method
of generating a SAR image on an spaceborne or aerial platform,
comprising receiving echoes of a chirp transmitted from a SAR
antenna; digitising the echoes into digital samples; carrying out
range compression on arrays of samples corresponding to range
lines; and carrying our azimuth compression on arrays of samples
corresponding to azimuth lines, wherein at least one out of
carrying out range compression and carrying out azimuth compression
comprises frequency demultiplexing signals corresponding to the
arrays into a plurality of channels, each signal corresponding to
an array comprising information about a plurality of target points
and each frequency channel comprising information about said
plurality of target points, and carrying out image formation on
each channel such that the range compression and the azimuth
compression generates a plurality of sub-images corresponding to a
plurality of frequency channels and wherein the method further
comprises combining said sub-images.
[0024] The frequency demultiplexing and the image formation may be
carried out in hardware.
[0025] The signal may be frequency demultiplexed before image
formation is carried out on each frequency channel.
[0026] Alternatively, the image formation on the channels may be
carried out before the channels are isolated and the frequency
demultiplexing may comprise frequency demultiplexing the signal
comprising the frequency channels in compressed form.
[0027] The range compression may comprise frequency demultiplexing
the range lines into a plurality of channels and range compressing
the channels individually using a compression filter to form a
plurality of intermediate matrices and the azimuth compression may
comprise frequency demultiplexing each azimuth line of each
intermediate matrix into a plurality of frequency channels and
azimuth compressing each frequency channel individually using a
compression filter to form the plurality of sub-images.
[0028] Alternatively, the range compression may comprise frequency
demultiplexing the range lines into a plurality of channels and
range compressing the channels individually using a compression
filter to form a plurality of intermediate matrices and the azimuth
compression may comprise averaging the samples of each azimuth line
or other unfocussed SAR processing to form the plurality of
sub-images.
[0029] Combining said images may comprise averaging the sub-images
to produce a compressed image.
[0030] Receiving echoes from an antenna may comprise receiving
echoes from a first antenna and combining the sub-images comprises
correlating corresponding sub-image from the first and the second
antenna to form a number of sub-interferograms and averaging the
sub-interferograms to generate a compressed interferogram.
BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Embodiments of the invention will now be described, by way
of example, with reference to the accompanying drawings, in
which:
[0032] FIG. 1 is a schematic diagram of a spacecraft payload
comprising a SAR antenna arrangement;
[0033] FIG. 2 is a schematic diagram of the components of the
digital processor of the SAR antenna arrangement;
[0034] FIG. 3 is a schematic diagram of a processor unit of the
digital processor shown in FIG. 2.
[0035] FIG. 4 is a schematic diagram of the sub-channel stacking
arrangement of a signal to be processed in the processor unit of
FIG. 3;
[0036] FIG. 5 is a functional diagram of a block channeliser of the
processor unit of FIG. 3;
[0037] FIG. 6 shows an alternative embodiment of a processor unit
of FIG. 2;
[0038] FIG. 7 illustrates a method for receiving and processing SAR
data;
[0039] FIG. 8 illustrates a method of processing a signal
corresponding to a range or azimuth line in a processor unit of
FIG. 3; and
[0040] FIG. 9 illustrates a method of processing a signal
corresponding to a range or azimuth line in a processor unit of
FIG. 6.
DETAILED DESCRIPTION
[0041] With reference to FIG. 1, there is provided a satellite
payload 1 comprising synthetic aperture radar (SAR) antenna
arrangement 2, a memory 3 and a transceiver 4 for receiving
commands and transmitting data to a control station (not shown) on
the ground. The satellite payload also comprises a controller 5 for
controlling the SAR antenna arrangement 2, the memory 3 and the
transceiver 4. The controller 5 may further receive commands from
the control station on ground. The SAR arrangement 2 can be used
for both transmitting radar pulses and receiving echoes. FIG. 1
shows the components in the receive chain. As shown in FIG. 1, the
SAR arrangement comprises an antenna 6, a low noise amplifier (LNA)
7, a downconverter 8 for down-converting the received SAR echoes to
baseband frequency, an analogue to digital converter (ADC) 9 and a
digital signal processor 10.
[0042] If the antenna 6 is also used to transmit the radar pulses,
arrangement 2 would also comprise a transmit chain with a digital
to analogue converter, an upconverter and a power amplifier. It
should be realised, that for sufficiently fast ADC and DAC
converters, the downconverter 8 and the upconverters may not be
used.
[0043] According to some embodiments, the satellite payload 1 may
include more than one SAR antenna, in which case there would be a
separate receive chain with a separate low noise amplifier,
downconverter and ADC for each antenna. If the antennas are also
used to transmit radar pulses, there would also be a separate
transmit chain for each antenna.
[0044] The antenna illuminates a wide area with a pulsed beam. Each
pulse may have a frequency that varies with time. The instantaneous
RF frequency of a pulse or "chirp" at time t is
f.sub.c+K.sub.chirpt, where f.sub.c is the centre or carrier
frequency and K.sub.chirp is the rate of change of the frequency of
the pulse or chirp. The maximum instantaneous frequency offset from
the carrier is K.sub.chirpT.sub.chirp, where T.sub.chirp is the
duration of the pulse. The bandwidth of the pulse, B.sub.chirp, can
be considered to be approximately equal to the maximum
instantaneous frequency offset:
B.sub.chirp.apprxeq.K.sub.chirpT.sub.chirp Equation 1
[0045] As an example, each chirp may have a bandwidth, B.sub.chirp,
of approximately 100 MHz.
[0046] The satellite travels in the along track or "azimuth"
direction and the pulses propagate in the across-track or "range"
direction. Each pulse illuminates a target area or a swath and, for
each transmitted pulse, the antenna arrangement 2 receives and
processes reflections from scatterers within the swath illuminated
by the pulse. The echoes received by the antenna 6 are sampled at a
frequency related to the bandwidth of the chirps. The samples of
the received echoes of a particular pulse forms a 1D array of
samples referred to as a "range line". In some embodiments, a
typical number of samples in each range line is 20,000. However, it
should be realised that the number of samples in a range line can
vary greatly with the application.
[0047] Each sample in the range line is associated with an index
which is proportional to the round-trip distance between the
antenna and the target. A range line comprising the sampled echoes
for a pulse is generated for each pulse forming a matrix of data
having one axis corresponding to the range direction and one axis
corresponding to the azimuth direction. A data array formed by the
samples that have the same index is referred to as an "azimuth
line". In some embodiments, a typical number of samples in an
azimuth line is 10,000. However, it should be realised that the
number of samples of the azimuth lines can vary depending on the
application.
[0048] The signal processor 10 processes the echoes of the chirps
according to an algorithm known as a "Range-Doppler" algorithm. The
Level 0 SAR data is processed in the signal processor 10 into Level
1 image data. The processing involves mapping points in the echo
matrix to corresponding points in a SAR image. In more detail, SAR
radar uses the fact that the distance between the antenna and a
target point changes with the relative movement. The signal energy
from a particular point of the target is spread in range and
azimuth and the SAR data processing collects this dispersed energy
into a single pixel in the output image. In range, the signal from
a particular point of the target is spread by the duration of
transmitted pulse and, in azimuth, the signal is spread by the
duration the target point is illuminated by the antenna beam.
[0049] With reference to FIG. 2, the signal processor 10 comprises
a pre-processor 11 for converting the signal into complex form, a
range processor 12 for image processing the data matrix in the
range direction, a range migration unit 13 for adjusting the data
in the matrix subsequent to the processing in the range direction,
an azimuth processor 14 for image processing the data matrix in the
azimuth direction and an image generator unit 15 for generating an
image.
[0050] The image processing of the data includes the comparison of
the echo signal with the original transmitted signals. This can be
accomplished with a convolution of the echo signals and the
transmitted pulses. Ideally, 2D convolutions in which every point
or pixel in the echoes matrix is mapped to its corresponding pixel
in the SAR image are carried out to create an image. However, this
would be very processing intensive and so in practice the range and
azimuth compression is typically decoupled into a set of 1D
convolutions on the range lines in the range processor 12, building
up a new intermediate matrix, and then a set of 1D convolutions on
the azimuth lines in the azimuth processor 14 to form a final
matrix corresponding to an image.
[0051] The range lines are provided to the range processor 12, via
the pre-processor 11, as they are digitised. The pre-processor 11
comprises a digital anti-aliasing filter (DAAF). The raw data of
each range line is input to the DAAF as a real digital signal and
the DAAF operates on the real incoming digital signal to convert it
into complex, baseband form, before passing the range line to the
range processor 12. The DAAF also decimates the sampling rate by a
factor of 2. The sampling frequency of the signal in the range
processor 12 and the azimuth processing unit will hereinafter be
denoted as f.sub.s. Accordingly, the sampling frequency of the
input signal to the DAAF would be 2f.sub.s.
[0052] The range processor 12 acts on each range line to range
compress the raw data. It carries out the 1D convolutions on each
range line and builds up one or more intermediate matrices in
memory 3, as will be described in more detail below.
[0053] The range compressed output is often skewed and once all the
range rows in the intermediate matrices are filled in, the range
migration unit 13 retrieves the intermediate matrices in the memory
and aligns the samples into the correct columns for the azimuth
processing. The range migration function can be implemented with a
suitable algorithm that acts on the samples in memory, as would be
realised by the skilled person in the art. The azimuth processor 14
then retrieves the adjusted one or more intermediate matrices from
the memory 3, carries out 1D convolutions on the azimuth lines as
will be described in more detail below and stores the final
matrices in the memory 3. The image generator unit 15 may carry out
further processing on the final matrices as will also be described
in more detail below.
[0054] Convolutions can be carried out using different techniques.
For example, direct convolution can be used. Moreover, fast
convolution with Fast Fourier Transform (FFT) techniques, or the
functionally equivalent so-called SPECAN (Spectral Analysis) method
can also be used.
[0055] According to embodiments of the invention, some or all
components of the digital processor are implemented in hardware.
The skilled person would realise that the DAAF and the range
migration function can easily be implemented in hardware. As will
be realised by the skilled person, the DAAF is a high speed, very
simple function which is ideally suited to hardware. The range
migration algorithm can be implemented in either hardware or
software depending on the application. According to the invention,
one or both of the range processing unit 12 and the azimuth
processor unit 14 is implemented using hardware, as will be
described in more detail with respect to FIG. 5. Additionally, in
some embodiments, the image generator unit 15 is also implemented
using hardware. In other embodiments, the image generator unit 15
may be a software function. The controller 5 may be a small general
purpose CPU with a control function implemented in software and one
or more of the components of the digital processor, but not all,
may be provided as part of the software control function.
[0056] With reference to FIG. 3, the components of a range
processor unit 12 or an azimuth processor unit 14 is shown. The
range processor unit 12 and the azimuth processor unit 14 have
corresponding components but components are configured differently
depending on whether they operate on range or azimuth lines. In the
description of FIG. 3 hereinafter we will refer to the processor
unit as a range processor unit and describe it with reference to
how it acts on a range line. However, it should be realised that
corresponding processing steps could be carried out for each
azimuth line in the azimuth processing unit but with a different
convolution function.
[0057] According to embodiments of the invention, the processor
unit 12, 14 provides more efficient processing by frequency
demultiplexing the data into a number K of sub-bands and creating a
sub-image for each sub-band before combining the sub-images. The
sub-bands will also be referred to as "frequency channels" herein.
Each frequency channel, demultiplexed from the full bandwidth
signal, contains information about all the scatterers in the target
area but to a lower resolution than the full bandwidth signal.
[0058] With reference to FIG. 3, the range processor arrangement is
implemented in hardware and comprises a block processor 16 which
provides the frequency channelisation and a decimation function and
a number of range compression filters 17a, 17k. By separating the
wide bandwidth into smaller channels, a direct convolution filter
can be used for each sub-channel which is readily implemented in
hardware.
[0059] The complex signal is provided from the DAAF 11 of the
signal processor to the block processor 16 where it is
demultiplexed into a number of sub-bands corresponding to different
frequency bands and where the sampling frequency for each sub-band
is reduced. The spectrum full bandwidth signals, which in the case
of a range line is the chirp bandwidth B.sub.chirp, is
schematically shown in FIG. 4 and can comprise a frequency
multiplex of K equally spaced sub-band channels in an even stacking
arrangement, centred at .omega.=2.pi.k/K where k=0, . . . K-1 and
where K is assumed even. In other words, for a range line, the
signal is demultiplexed into a plurality of frequency sub-bands of
the full chirp bandwidth. K may, for example, be equal to 100.
However, it should be understood that K may have any suitable
value. The block channeliser 16, which will be described in more
detail with respect to FIG. 5, is a computationally efficient,
FFT-based filter processor which divides the input into its K
constituent sub-band channels. The sampling rate of each sub-band
is reduced by a factor of M in the block channeliser, where M is
smaller or equal to the number of channels. It should be noted that
when M=K, maximum decimation is provided and the sub-bands are
critically sampled.
[0060] Each sub-band channel is then provided with a sampling rate
of fs/M to a separate compression filter 17a-17k which carries out
a direct convolution of each sub-band. In other words, the
compression of the signal is carried out on each sub-band
separately.
[0061] It will be appreciated that the full bandwidth signal
corresponding to a range line comprises information about a
plurality of target points and each frequency sub-band will
comprise information about each of the plurality of target points.
In other words, each frequency sub-band will comprise information
about the same target points but with a lower resolution than the
full bandwidth signal.
[0062] By carrying out range or azimuth compression on a number of
sub-bands instead of on the whole bandwidth, the processing rate of
the filters 17a-17k can be reduced. Denoting the length of the full
bandwidth compression filter by N, the computation rate in a direct
convolution of the input with this filter is Nf.sub.s complex
multiplications per second.
Computation rate.sub.no channelisation.ltoreq.Nf.sub.s Equation
2
[0063] In some embodiments, the compression filters 17a-17k are
identical impulse response filters, each operating on their
respective sub-channel. The filter impulse response of each filter
is the complex conjugate of the transmitted chirp waveform and the
length of the filter N is the number of chirp waveform samples in
the chirp period T.sub.chirp. Consequently, when the transmitted
chirp waveform is sampled at the full rate f.sub.s, we have:
N=T.sub.chirp/(1/f.sub.s). Equation 3
[0064] The full bandwidth compression filter for range processing
spans a bandwidth B.sub.chirp which can be expressed as
[0065] B.sub.chirp.apprxeq.K.sub.chirpT.sub.chirp, as mentioned
with respect to Equation 1 above.
[0066] The critically sampled sub-band has bandwidth B.sub.chirp/K.
Consequently, the time it takes it takes for the chirp at rate
K.sub.chirp to sweep across the sub-band T.sub.sub-band, chirp is
T.sub.chirp/K seconds. Moreover, since the sub-band has a lower
bandwidth, the sampling rate can be reduced to f.sub.s/M, as shown
in FIG. 3, where M.ltoreq.K. Consequently, the sampling time step
in the sub-band channels is M/f, seconds. As should be realised by
the skilled person, when M=K, the sampling rate is matched to the
bandwidth and the sub-band is critically sampled. Conversely, if
M<K, for example K/2, the sub-band is oversampled.
[0067] Denoting the length of the sub-band filter compression
filter as N.sub.sub, it follows that the length of the sub-band
filter equals the time it takes for the chirp to sweep across the
sub-band divided by the sampling time step:
N sub = T sub - band , chirp f s M = T chirp K f s M Equation 4
##EQU00002##
[0068] Using Equation 3, this becomes
N sub = N KM Equation 5 ##EQU00003##
[0069] When the sub-band is critically sampled, M=K and so
N.sub.sub=N/K.sup.2. When the sub-band is oversampled, e.g. by a
factor of 2 with M=K/2, then the filter is longer, e.g.
N.sub.sub=2N/K.sup.2.
[0070] The computational rate in the bank of sub-band compression
filters is the number of filters K, multiplied by the rate of each
sub-band f.sub.s/M and the length of each filter N.sub.sub,
computation
rate.sub.channelised=K(f.sub.s/M)N.sub.sub=K(f.sub.s/M)N/(MK)=Nf.sub.s/M.-
sup.2 Equation 6
[0071] For the critically sampled case, the computational rate can
be expressed as,
computation rate.sub.channelised=Nf.sub.s/K.sup.2 Equation 7
[0072] Consequently, it can be seen that the convolution
computational load in the compression filters acting on the
sub-bands is a factor of 1/K.sup.2 of that in the direct
convolution of the whole bandwidth signal when M=K.
[0073] When all range lines have been processed, a full
intermediate matrix for each channel has been generated in memory
3. Once the matrices have been adjusted in the range migration unit
13, each azimuth line of an intermediate matrix is fed to the
azimuth processor 14 to be individually processed. The intermediate
matrices generated by the range compressor may be processed
sequentially. As mentioned above, the azimuth processor 14 can also
be implemented as shown in FIG. 3. Each azimuth line of each
intermediate matrix can be channelised into K' channels and the
azimuth compression to be carried out on each of the K' channels
individually creating K' new matrices for each intermediate matrix.
In the azimuth processing unit, the block channeliser 16 will
typically divide the data into a different number of channels and
the compression filter has different coefficients corresponding to
a different impulse response. Moreover, in the azimuth processing
unit, the frequency bandwidth is the Doppler bandwidth and not the
bandwidth of each pulse. As the SAR antenna moves over a target
point, the distance to the target point changes which causes a
phase variation in the received signal as a function of azimuth.
The variation in phase between adjacent samples in an azimuth line
depends on the pulse repetition rate and how quickly the antenna
moves over the target. The bandwidth in azimuth, which results from
the phase variation, is known as the Doppler bandwidth.
Accordingly, for an azimuth line, the signal is demultiplexed into
a plurality of frequency sub-bands of the Doppler bandwidth of the
received echoes. It will be appreciated that the full bandwidth
signal corresponding to an azimuth line comprises information about
a plurality of target points and each frequency sub-band to which
the signal is demultiplexed will comprise information about each of
the plurality of target points. In other words, each frequency
sub-band will comprise information about the same target points but
with a lower resolution than the full bandwidth signal.
[0074] The full azimuth bandwidth compression filter would also
have the form of a chirp waveform, just like in range, but with a
different centre frequency and chirp rate. Consequently, the
computational savings in performing azimuth compression on
frequency channels as opposed to direct convolution on the full
bandwidth signal is also proportional to 1/K'.sup.2, where K' is
the number of channels into which the azimuth bandwidth is divided.
However, it should be realised that typically the azimuth bandwidth
is much smaller than the range bandwidth and the number of channels
used during azimuth processing is typically much smaller than the
number of channels used during range processing.
[0075] When the range compression and the azimuth compression have
been completed, the initial raw data matrix has been transformed
into a number of sub-images in the memory 3. All the sub-images are
of the same target area. In the embodiments where both the range
processor unit and the azimuth processor unit divides the data into
frequency channels, the number of sub-images is KK' where K is the
number of channels used during the range processing and K' is the
number of channels used during the azimuth processing.
[0076] It should be realised that the sub-images are fully formed
SAR images in their own right, just at a lower resolution than the
image that would have been obtained had the raw data been processed
at its full bandwidth.
[0077] With reference to FIG. 2 again, the image generator 15 forms
a SAR image based on the sub-images. In some embodiments, the image
generator 15 is configured to combine the sub-images into a full
bandwidth output signal. However, in other embodiments, the image
generator 15 is configured to combine the sub-images by averaging
in order to produce a low-resolution image with an improved signal
to noise ratio compared to the sub-images. The images can be
averaged by simply stacking the images on top of each other and
adding corresponding pixels. If KK'=100, the amount of data that is
required to be stored can be reduced by a factor of 100 by
averaging the sub-images instead of adding the images together to
obtain a full resolution image. In some embodiments, a subset of
the images is added together and different subsets are then
averaged to provide a better resolution image. A significantly
smaller volume of storage would be required to store the averaged
Level 1 product compared to the volume of storage required to store
the Level 0 data from which the averaged Level 1 product is
obtained.
[0078] In some embodiments, the image generator can further be
configured to generate interferograms as part of the Level 1
product. If the spacecraft has two or more SAR antennas, the
sub-images from the different antennas can be correlated in the
image generator 15 to produce a number of sub-interferograms. An
interferogram with an improved signal to noise ratio can then be
achieved by averaging the sub-interferograms.
[0079] Moreover, in some embodiments, the image generator may apply
standard image compression techniques to the averaged Level 1
product to further reduce the data volume. As an example, jpeg
compression techniques can be used to further compress the data.
However, it should be realised that any suitable image compression
technique can be used. The averaged SAR images or averaged
interferograms can be further reduced using standard image
compression techniques prior to storage and/or downlink
transmission.
[0080] A suitable architecture for the block processor 16 of FIG. 3
is shown with respect to FIG. 5. In brief, the block processor 16
comprises block fine processing stages 18a, 18b and block coarse
processing stages 19a, 19b. The block processor and block
processing stages are a "block" processor and "block" processing
stages in the sense that they apply a single computational function
simultaneously to all the frequency channels in the input signal.
The fine processing stage 18a, 18b and the block coarse processing
stage 19a, 19b together demultiplex the signal corresponding to a
range line or an azimuth line into a plurality of frequency
channels. In the embodiment shown in FIG. 5, the block fine
processing stages are provided by two imaged half-band filters 18a,
18b configured to generate from the signal corresponding to a range
line or an azimuth line a frequency division multiplex (FDM) signal
with a number of even index frequency channels and an FDM signal
with a number of odd index frequency channels respectively. The
region of the spectrum corresponding to the odd index channels has
been nulled in the signal with the even index channels and the
region of the spectrum corresponding to the even index channels has
been nulled in signal with the odd index channels. The half-band
filters of the block channeliser may comprise one high pass filter
and one low pass filter. The impulse response of the half-band
filters are padded with N-1 zeros to produce N even numbered and N
odd numbered images of the basic, tight prototype frequency
response of which the images are copies.
[0081] The block coarse processing stages 19a, 19b then isolate the
even numbered and the odd numbered channels from the FDM signals
output by the fine processing stages. In more detail, each block
processing stage comprises a Weighted OverLap Add (WOLA) unit 20a,
20b and a linked K/2 point FFT unit 21a, 21b. Each FDM signal from
the fine processing stages is provided to a separate WOLA units
20a, 20b and its linked K/2 point FFT units 20a, 20b. Each WOLA
unit and associated K/2-point FFT unit generate a plurality of
narrow band channels from their respective FDM signal. Each WOLA
performs a filtering function with a response having transition
bands in the nulled out regions of the spectrum of the associated
FDM signal. In some embodiments, each WOLA unit and linked
K/2-point FFT unit may be replaced with a polyphase-Discrete
Fourier Transform (DFT) channeliser. Decimation by a factor of M is
carried out within the block channeliser 16. Conceptually, the
block channelisation occurs at the full sampling rate of the input
and each sub-band output is oversampled by the factor M. The
sub-band channel outputs can then be decimated by a factor of M
without causing aliasing. In practice, the decimation factor is
absorbed into the block channeliser structure so that the unwanted
samples in the outputs are simply not computed, achieving
computational savings.
[0082] The block signal generator can be incorporated in an
application specific integrated circuit (ASIC) or a field
programmable gate array (FPGA). A more detailed description of a
suitable architecture for the block processor, together with
modifications and variations, can be found in EP0695054 and
EP0831611. However, it should be realised that any suitable
architecture for the block channeliser can be used.
[0083] Another embodiment of the range processor unit 12 or the
azimuth processor unit 14 is shown in FIG. 6. In FIGS. 3 and 6,
like reference numerals designate like components. However, in FIG.
6, instead of providing separate compression filters for the
channels, a compression filter 17 with an imaged frequency response
is provided to operate on the signal before being provided to the
block processor 16. The compression filter 17 is derived from the
prototype filters 17a-17k of FIG. 3. The compression filter
response is interpolated by padding K-1 zero samples between each
input sample to form the frequency response
H'(e.sup.i.omega.)=H(e.sup.i.omega.K), where H(e.sup.i.omega.) is
the frequency response of the prototype filter h(n) and
h ' ( n ) = { h ( n / K ) , n = 0 , .+-. K , .+-. 2 K 0 , otherwise
##EQU00004##
[0084] The imaged filter therefore does not act on the whole
bandwidth but performs low resolution compression on the individual
channels as a block process acting on the complete wideband signal.
The subsequent block channeliser 16 then frequency demultiplexes
the sub-bands in already compressed form. In other words, the
compression can be considered to be carried out on "conceptual"
frequency channels before the frequency channels are isolated. The
imaged compression filter still has N.sub.sub non-zero coefficients
for each sub-channel, with N.sub.sub calculated for the critically
sampled case, so the computational workload for the arrangement of
FIG. 6 is f.sub.sN.sub.sub=f.sub.8N/K.sup.2, which is the same as
the arrangement of FIG. 3 for the critically sampled case, as shown
in Equation 7.
[0085] The digital processor of FIG. 6 provides a reduced
computational load when the sub-bands are oversampled (e.g. by 2)
to aid subsequent SAR processing. For example, if a number of
different SAR antennas were used, oversampling may be required to
co-register and interpolate corresponding sub-images prior to
forming an interferogram or a complex coherence map. In this case
the workload in the bank of compression filters of the first
embodiment is greater than the workload in the single compression
filter of the second embodiment, as 1/K.sup.2 versus 1/M.sup.2. For
example if M=K/2, the block compression in the second embodiment is
4 times more efficient (1/K.sup.2 versus 4/K.sup.2).
[0086] With reference to FIG. 7, a method of receiving and
processing SAR signals is provided. The echoes from different
chirps are received in the antenna 6 at step S7.1. The echoes are
then sampled at step S7.2. Immediately after the signals are
digitised they are passed to the DAAF filter and processed into
complex form at step S7.3. As the range lines come in and are
converted into complex form, they are passed to the range processor
unit 12 and range compressed at step S7.4. As the range lines are
processed at step S7.4, the rows in the intermediate matrices are
filled to form a number of full intermediate matrices. It should be
realised that individual steps of S7.1, S7.2, S7.3 and S7.4 may not
be finished before the next step is begun. The range processor may
be processing one range line as the echoes for another range line
for the same image are received.
[0087] At step S7.5, the data in the matrix is adjusted and moved
into the right columns for the azimuth processing and the azimuth
lines of the intermediate matrices are then azimuth compressed at
step S7.6 to form a number of sub-image matrices. At step S7.7, the
sub-images may be averaged to form a final compressed
low-resolution image. The compressed image can further processed
and/or stored until a suitable time when it can be sent to a ground
station.
[0088] Steps S7.4 or S7.6 are described in more detail with respect
to FIGS. 8 and 9. It should be realised that in some embodiments,
the type of processing unit shown in FIG. 3 or 6, which divides the
signal into a number of channels, is used for only the range
compression or only the azimuth compression and one of the range
compression and the azimuth compression is carried out on the full
bandwidth or using other alternative techniques. For example,
alternative techniques would be appropriate for the azimuth
compression in unfocussed SAR. However, in other embodiments, the
signal corresponding to an array of data is divided into frequency
channels during both the range compression and the azimuth
compression. With respect to FIGS. 8 and 9, the data signal,
corresponding to either a range line or an azimuth line, is either
demultiplexed into channels at step S8.1 for the compression to be
carried out by a number of compression filters on individual
channels S8.2 in an processing arrangement shown in FIG. 3 or
channels of the signal are compressed at step S9.1 before the
compressed channels are demultiplexed at step S9.2 in a processing
arrangement as shown in FIG. 6.
[0089] Whilst specific examples of the invention have been
described, the scope of the invention is defined by the appended
claims and not limited to the examples. The invention could
therefore be implemented in other ways, as would be appreciated by
those skilled in the art.
[0090] For example, as described above, only one of the range
compression and the azimuth compression may be implemented as
described with respect to FIG. 3 and FIG. 6. It is contemplated
that the processing technique and the processing unit described
with respect to FIGS. 3 and 6 may only be used for the range line
and the compression in azimuth is obtained by averaging the samples
in each azimuth line or by an alternative technique appropriate for
an unfocussed SAR system. In some embodiments, the range
compression is implemented as described with respect to FIG. 3 or
FIG. 6 in hardware and the azimuth compression is implemented in
software. Since the bandwidths involved in azimuth processing is
generally so much lower, the processing rates are sufficiently slow
for the software on a space CPU to be able to carry out the azimuth
processing in some cases. Consequently, in applications where the
azimuth processing is simplified, the azimuth processing may be
carried out in software.
[0091] It should be realised that although the components of the
signal processor 10 have been described as implemented in hardware
or software, some of the functions provided by the components can
be implemented in a combination of hardware and software.
[0092] Moreover, it should be realised that the invention is not
restricted to satellites and other space platforms. The invention
could be used in any suitable application, especially in
applications where processor mass, power and particularly TM
bandwidth are at a premium. For example, the invention could be
used in an unmanned aerial vehicle (UAV) or other aerial platform.
For UAV, the real-time imagery aspect of the invention would be
particularly relevant.
[0093] Additionally, it should be realised that although it has
been described with respect to FIG. 2 that the range lines are
provided directly from the pre-processor 11 to the range processor
unit 12, the range processor unit 12 does not have to act in
real-time on the samples from the ADC 9. Instead, the ADC output
could be stored in the memory for processing at a later time. Such
an approach would be useful in missions with a relatively short
data acquisition and storage phase, followed at leisure by a
processing stage to reduce the data volume before downlink
transmission. An example of such a mission would be a planetary
flyby mission.
[0094] Also, it should be realised that although specific examples
of the block channeliser and the compression filters have been
described, any suitable architectures can be used for frequency
demultiplexing the signals into channels and carrying out image
formation on the channels individually. Moreover, the number and
order of compression filters and channelisers can be varied
depending on the application. It is contemplated that although a
single imaged response filter has been described with reference to
FIG. 6, a number of imaged response filters may be provided to
carry out image formation on the channels separately before the
channels are isolated.
* * * * *